Análise de componentes principais em séries temporais multivariadas com heteroscedasticidade condicional e outliers : uma aplicação para a poluição do ar, na Região da Grande Vitória, Espírito Santo, Brasil

• Ano de defesa: 2016
• Autor(a) principal: Monte, Edson Zambon
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Espírito Santo; BR; Doutorado em Engenharia Ambiental; Centro Tecnológico; UFES; Programa de Pós-Graduação em Engenharia Ambiental
• Palavras-chave em Português: Principal Component Analysis.; Conditional Heteroscedasticity.; Outliers.; Air Pollution.; Robustness.; Analise de Componentes Principais; Heteroscedasticidade Condicional.; Outliers; Heteroscedasticidade (Estatística); Robustez; Poluição do Ar; Análise de componentes principais.; Valores estranhos (Estatistica).; Estatística robusta.; Ar - Poluição.; Vitória, Região Metropolitana de (ES).; Engenharia Sanitária; 628
• Link de acesso: http://repositorio.ufes.br/handle/10/10332

Issues relating to air quality have become increasingly important, since many health problems come from air pollution. In addition, air pollution contributes to the degradation of the environment, contributing to the greenhouse effect. Thus, several studies adopting technical statistics have been conducted in order to contribute in the making of public and private actors with regard to combating pollution, prevention of high concentrations and formulation of laws for this purpose. The classical principal component analysis (PCA) is a statistical methodologies adopted. The PCA is used for dimensional reduction, cluster analysis, regression analysis, among others. However, among the studies that have adopted the classical PCA, a common feature is to neglect the conditional heteroscedasticity and/or the presence of additive outliers, which may lead to spurious results (misleading), since the estimated autocovariance matrix may be biased (estimated incorrectly). It is possible to note that the time series related to air pollution tend to present conditional heteroscedasticity and additive outliers. Then, the first paper of this thesis proposed to apply a multivariate filter VARFIMA-GARCH to the original data and use the classical PCA on residuals of the VARFIMA-GARCH model. Besides the volatility, this model was used to filter the temporal correlation and the long memory behavior. The application of the PCA on the residuals of the VARFIMA-GARCH model was more consistent with the environmental characteristics of the Greater Victoria Region (GVR), Esp´ırito Santo, Brazil, than the application using the original data The second paper, that is the core of this thesis, the technique of principal volatility components (PVC), proposed by Hu e Tsay (2014), was extended for a robust approach (RPVC), in order to capture the volatility present in the multivariate time processes, but considering the effects of additive outliers on conditional covariance, since these outliers may mask (“hide”) the conditional heteroscedasticity or even produce spurious volatility. The proposed RPVC improved the predictions of PM10 exceedance days in the Laranjeiras station, in the GVR.